Multi-Target Custody via Distributed Multiple-Model Multiple-Hypothesis Tracking

The challenge of multi-target tracking is made more onerous when there are nontrivial sensor coverage gaps. Most automatic trackers will drop track outside of coverage, due to the large track state uncertainty. Recently, use of the Ornstein-Uhlenbeck (OU) process has been proposed to extend object custody outside regions of sensor coverage, by exploiting context information that supplements available sensor data. This paper continues this line of research using a multiple-model OU filter within a distributed multiple-hypothesis tracking paradigm, with local (context-unaware) trackers and downstream (context-aware) track fusion. Crucially, the context-aware tracking filter must allow for objects that deviate from expected behavior; we achieve this via a context-unaware mode with limited interaction with context-aware modes. Further, as context may include multiple competing modes, it is necessary to generalize existing tracking performance metrics to account for partial object custody. Accordingly, we introduce multiple-model extensions to target completeness, track completeness, and optimal sub-pattern assignment (OSPA) metrics. We quantify improved performance of context-aware surveillance over baseline processing using variations of an air surveillance scenario.
^1
1
IEEE Aerospace Conference, Big Sky MT, March 2022.
^2
2
IEEEAC paper #2705, Final Version, Updated 2021-11-24.